The meeting notes graveyard
Another steering call ends. Someone pastes bullets into Slack. Slides land in a folder named Final_v7_REALLY_FINAL. Three weeks later, nobody remembers what "Option B" meant—and the AI assistant cannot help because knowledge never entered the system of record.
AI meeting summarization saves time only if summaries land where work lives. WKFGo provides publish_meeting_wiki via MCP—a workflow to turn meeting records into searchable wiki pages linked to projects and tasks.
What a publishable meeting record includes
Before AI summarizes, capture structure humans (and tools) can rely on:
| Section | Purpose |
|---|---|
| Context | Project, attendees, date |
| Decisions | Explicit forks chosen—not vague "aligned" |
| Open questions | Unresolved items with owners |
| Links | Tasks, finance entries, releases referenced |
| Next meeting | What evidence should exist by then |
AI drafts sections from transcript or notes; facilitator edits before publish.
publish_meeting_wiki workflow
- Capture — Notes or transcript from the call
- Draft — AI summarizes decisions vs discussion vs actions
- Review — Human fixes names, dates, and decision wording
- Publish —
publish_meeting_wikicreates/updates wiki page - Discover — Team uses
smart_searchand in-app wiki; MCP clients query same index
Published pages survive turnover—new hires find "why we descoped Feature X" without archaeology in email.
AI summarization principles
Decisions vs conversation — Models excel at narrative; humans must mark Decision: lines explicitly so summaries do not soften commitments.
No invented action items — AI proposes; humans confirm owners before create_task (companion workflow).
Bilingual teams — WKFGo supports English/Persian UI; summarize in the language the wiki page uses for consistency.
Permission-aware — Sensitive steering content publishes to project wiki with same access rules—not public marketing site.
Rituals that make publishing stick
- Template recurring meetings — Sprint review, steering, retro—same headings every time
- Publish within 24 hours — Memory decays; search value drops
- Link from tasks — Reference wiki slug in related task comments
- Retro metric — "Decisions published?" checkbox beats "had a retro"
Integration with executive workflows
Leadership meetings feed decision_inbox and executive_brief—wiki publish is the durable narrative those tools reference next cycle. When CFO asks "why did we delay Release 2?", smart_search finds the steering wiki—not a vague memory.
Anti-patterns
- Summaries only in chat threads
- AI publish without human review on client-facing decisions
- Wiki pages titled "Meeting 3/12" with no project context
- Skipping
publish_meeting_wikibecause "we'll clean it up later"
Transcription and privacy boundaries
If you transcribe calls externally, confirm retention policy before feeding text to models. publish_meeting_wiki should include only content appropriate for project wiki permissions—redact client secrets and personnel matters in human review, not in automated publish.
For internal standups, shorter notes often beat verbatim transcript: decisions and blockers matter more than dialogue polish.
Search after publish: proving value
One week after publish, run smart_search for a decision phrase from the meeting. If the wiki page ranks first, knowledge succeeded. If not, improve title and headings—search quality reflects writing habits, not only AI summary quality.
Meeting types and summary depth
| Meeting | Summary depth |
|---|---|
| Daily standup | Blockers + dates only—often no wiki publish |
| Sprint review | Decisions + demo notes—publish |
| Steering | Decisions + budget/release—publish + inbox |
| Retro | Process actions—publish; create_task for fixes |
AI should not apply steering-level verbosity to fifteen-minute standups—match format to ceremony.
Attendees list in the wiki page
Include an Attendees section in every published page — it cuts off the later "I wasn't in that call" excuse and tells a new hire exactly whose context to pull in. AI can draft the list from the calendar invite; a human still fixes names before publish.
Meeting summaries die in chat threads. Standardize: within twenty-four hours of a decision meeting, publish wiki notes and link related tasks. publish_meeting_wiki makes summaries searchable for smart_search and MCP—so the next 'what did we decide?' question cites a page, not memory.
Run a monthly audit: do AI answers cite task IDs from MCP? If not, tighten prompts to require citations before any stakeholder-facing draft leaves the team.
Permission-aware tools mean the same question gets different grounded answers for PM vs contractor—that is a feature, not a bug. Train teams to expect scoped truth.
Decision hygiene compounds: log outcomes in decision_log, compare forecasts to results next quarter, and adjust thresholds—not just models.
FAQ — AI meeting summarization
Does publish_meeting_wiki replace the chatbot?
No—it feeds knowledge the chatbot and smart_search retrieve.
Can AI summarize from audio?
Transcription depends on your toolchain; WKFGo publish step accepts reviewed text.
Who can publish?
Wiki edit permissions apply—same as manual wiki updates.
How do assistants use published notes?
MCP get_wiki_page and smart_search query published content with citations.
Try it now
Put these patterns on live project data—not slide decks.